Higher runtimes on management IPs

[Freihafener]

Hi,

I have some trouble in a LAN environment. A ping on the management IP of the core has a 1-6ms latency, the other switches 8-30ms. From time to time first pings from a client to the story switches get a timeout or really high latency.

A ping from a connected client to another has an expected latency of <1ms.

Anyone any idea where to look at?

Our infrastructure at this office for short (former server room):

About  ten application servers (GigabitEth) and 300 clients (ThinClients  without VOIP), a 4506 core / WAN Gateway connecting 14 switches in the stories, one VLAN for all.

Greetz FH

[Freihafener]

BTW: looked at the CPU utilization of the 4506. the average utilization  is still at 10-20%, but the max is at 60-70% very often... strange in  such a little networt, isn´t it?!

[Jon Marshall]

It's important to realise that a ping to the switch is handled in software whereas a ping from a client to another client through the switch is handled in hardware.

So the difference in times you see is explained purely by how busy the switch is in software at any one time. If you are getting timeouts it may simply mean the switch is too busy to repsond in time to the ping. You may want to investigate exactly what is pushing the CPU up to 60/70%.

But the key thing to note is that a relatvely slow response from the switch itself does not necessarily mean that user traffic through the switch is being affected. User traffic will be only be software switched under very specific conditions eg. on some L3 switches a "log" option in acls can send the traffic to the main CPU (ie. software switched), deny statements in acls attached to policy routing can also do this.

Jon

[Freihafener]

Dear Jon,

thanks for your reply. I know about the difference between sw an hw switching. I must confess I have not checked the cpu load of the switches on the stories yet -  maybe they have that feature, maybe not.

As I mentioned the client communication is not affected - but I wonder how 300 thin clients and a bunch of servers can make the switches busy.

FH

[Jon Marshall]

It really does depend on the switch config. As i say, most user traffic is hardware switches but certain things in your config can mean the traffic is sent to the main CPU ie. software switched.

You could also consider non user traffic affecting the switches ie. STP issues. When the switch hits 60/70% do you know what is using most of the CPU ?

Jon

[Freihafener]

On the 4506 core we have a cpu utilization of 5% from

Cat4k Mgmt HiPri

Cat4k Mgmt LoPri

As you mentioned STP - we have two switches on the stories which bring a STP state change for any client coming up or down: %SPANTREE-6-PORT_STATE.

For that matter I got about 800 messages a day concerning STP change. Do you mean that could be the reason? Could deactivate spanning tree for the client ports - but what if a foxy user connects a SOHO switch in the office the wrong way...?

[Jon Marshall · Accepted solution]

No. don't disable STP on client ports. Make sure they are portfast but don't disable. No problem with messages as these are normal when a client comes on line or shuts down.

To be honest, if the client communication is unaffected i wouldn't worry too much about ping times to and from switch itself.

Jon

[Freihafener]

Guess you are right - but I am sure that the response times were better in the past. So I had the apprehension that slow ping answers could be the beginning of something bad affecting the clients.

But probably if you cannot find any reason for a problem there is no problem at all.

Thanks for your estimation.

FH